Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. An object recognition apparatus comprising: a heterogeneous property data generation unit that generates, for a plurality of different properties, property data which highlights a specific property based on target data acquired for an object; a discrimination-use feature amount extraction unit that extracts, for each piece of the property data, a discrimination-use feature amount used for discrimination of each piece of the property data; a discrimination information calculation unit that calculates, for each piece of the property data, discrimination information used for discrimination of the property data based on the discrimination-use feature amount calculated for each piece of the property data; a reliability feature amount extraction unit that extracts, for each piece of the property data, a reliability feature amount used for estimation of reliability of the discrimination information calculated for each piece of the property data; a reliability estimation unit that estimates, for each piece of the property data, the reliability of the discrimination information based on the reliability feature amount calculated for each piece of the property data; a synthesized information generation unit that generates synthesized information acquired by synthesizing the discrimination information calculated for each piece of the property data and the reliability calculated for each piece of the property data; and an object recognition processing unit that performs processing related to recognition of the object based on the synthesized information and generates a recognition result.
2. The object recognition apparatus according to claim 1 , further comprising: an information storage unit that stores discriminator information for construction of a discriminator prepared for each piece of the property data, wherein the discrimination information calculation unit calculates the discrimination information on the property data based on the discrimination-use feature amount of the property data and the discriminator prepared for the property data.
3. The object recognition apparatus according to claim 2 , further comprising: an information storage unit that stores reliability estimator information for construction of a correct discrimination reliability estimator and an incorrect discrimination reliability estimator which are prepared for each piece of the property data, wherein the reliability estimation unit includes: a correct discrimination likelihood value calculation unit that calculates a correct discrimination likelihood value based on the reliability feature amount of the property data and the correct discrimination reliability estimator of the property data; an incorrect discrimination likelihood value calculation unit that calculates an incorrect discrimination likelihood value based on the reliability feature amount of the property data and the incorrect discrimination reliability estimator of the property data; and a reliability calculation unit that calculates the reliability based on the correct discrimination likelihood value and the incorrect discrimination likelihood value.
4. The object recognition apparatus according to claim 3 , wherein the information storage unit stores an evaluation value of the discriminator, and the reliability calculation unit calculates the reliability based on the correct discrimination likelihood value, the incorrect discrimination likelihood value, and the evaluation value.
5. The object recognition apparatus according to claim 4 , wherein the information storage unit stores discriminator weight information on a weight of each of the discriminator, and the synthesized information generation unit generates synthesized vector information based on the discrimination information, the reliability, and the weight of the discriminator as the synthesized information.
6. The object recognition apparatus according to claim 5 , wherein the object recognition processing unit generates a class estimated based on the synthesized vector information as the recognition result.
7. The object recognition apparatus according to claim 6 , further comprising: an information storage unit that stores a correspondence between learning-use discrimination-use feature amount and the class as learning-use information; and a discriminator construction unit that generates the discriminator information based on the learning-use discrimination-use feature amount.
8. The object recognition apparatus according to claim 6 , further comprising: an correctness or incorrectness evaluation unit that generates a correctness or incorrectness attached reliability feature amount by estimating a class based on learning-use discrimination information and comparing the estimated class with a class corresponding to the learning-use discrimination-use feature amount, the correctness or incorrectness attached reliability feature amount being information including a result of correctness or incorrectness evaluation for each piece of the property data; and a discriminator weight information generation unit that generates the discriminator weight information based on a ratio of correctness to incorrectness grasped from the correctness or incorrectness attached reliability feature amount, wherein the information storage unit stores a correspondence among the learning-use discrimination-use feature amount, learning-use reliability feature amount, and the class as learning-use information, and the discrimination information calculation unit generates the learning-use discrimination information based on the learning-use discrimination-use feature amount.
9. The object recognition apparatus according to claim 8 , further comprising: an reliability estimator construction unit that includes a correct discrimination reliability estimator construction unit which constructs the correct discrimination reliability estimator based on the correctness or incorrectness attached reliability feature amount, an incorrect discrimination reliability estimator construction unit which constructs the incorrect discrimination reliability estimator based on the correctness or incorrectness attached reliability feature amount, and an evaluation value calculation unit which calculates the evaluation value based on the correctness or incorrectness attached reliability feature amount, and that generates the reliability estimator information based on the correct discrimination reliability estimator, the incorrect discrimination reliability estimator and the evaluation value.
10. The object recognition apparatus according to claim 7 , further comprising: a class setting unit that provides a user interface which sets the learning-use information.
11. The object recognition apparatus according to claim 1 , wherein the target data is image data in a Red Green Blue (RGB) format acquired by photographing the object with a photographing apparatus, and the heterogeneous property data generation unit generates at least two of data acquired by extracting an R component from the image data, data acquired by extracting a G component from the image data, and data acquired by extracting a B component from the image data as a plurality of pieces of property data.
12. The object recognition apparatus according to claim 1 , wherein the target data is X-ray image data acquired by photographing the object with an X-ray photographing apparatus, and the heterogeneous property data generation unit is configured to generate a grayscale image having a pixel value acquired by scaling an X-ray intensity value in the X-ray image data as first property data, generate image data in a Red Green Blue (RGB) format converted from the X-ray image data in accordance with a range of the X-ray intensity value in the X-ray image data as second property data, and generate image data in a Hue Saturation Value (HSV) format converted from the X-ray image data in accordance with the range of the X-ray intensity value in the X-ray image data as third property data.
This invention relates to an object recognition apparatus that processes X-ray image data to enhance recognition accuracy. The apparatus addresses the challenge of effectively analyzing X-ray images, which often contain subtle variations in intensity that are difficult to interpret using conventional methods. The system generates multiple heterogeneous property data representations from the same X-ray image to improve feature extraction and recognition performance. The apparatus includes a heterogeneous property data generation unit that processes X-ray image data acquired by an X-ray photographing apparatus. It generates three distinct types of image data from the X-ray intensity values. First, it creates a grayscale image where pixel values are scaled versions of the X-ray intensity values. Second, it converts the X-ray image into an RGB format, adjusting the conversion based on the intensity range of the original X-ray data. Third, it converts the X-ray image into an HSV format, also scaled according to the intensity range. These multiple representations provide complementary information, enhancing the ability to detect and recognize objects within the X-ray images. The apparatus leverages these diverse data formats to improve the robustness and accuracy of object recognition tasks in X-ray imaging applications.
13. An object recognition system that includes the object recognition apparatus according to claim 6 , the object recognition system comprising: a learning apparatus communicably connected to the object recognition apparatus, wherein the learning apparatus includes: an information storage unit that storages a correspondence among learning-use discrimination-use feature amount, learning-use reliability feature amount, and the class as learning-use information; a discriminator construction unit that generates the discriminator information based on the learning-use discrimination-use feature amount; an correctness or incorrectness evaluation unit that generates a correctness or incorrectness attached reliability feature amount by (i) estimating a class based on learning-use discrimination information generated by the discrimination information calculation unit based on the learning-use discrimination-use feature amount and (ii) comparing the estimated class with a class corresponding to the learning-use discrimination-use feature amount, the correctness or incorrectness attached reliability feature amount being information including a result of correctness or incorrectness evaluation for each piece of the property data; a discriminator weight information generation unit that generates the discriminator weight information based on a ratio of correctness to incorrectness grasped from the correctness or incorrectness attached reliability feature amount; and an reliability estimator construction unit that includes a correct discrimination reliability estimator construction unit which constructs the correct discrimination reliability estimator based on the correctness or incorrectness attached reliability feature amount, an incorrect discrimination reliability estimator construction unit which constructs the incorrect discrimination reliability estimator based on the correctness or incorrectness attached reliability feature amount, and an evaluation value calculation unit which calculates the evaluation value based on the correctness or incorrectness attached reliability feature amount, and that generates the reliability estimator information based on the correct discrimination reliability estimator, the incorrect discrimination reliability estimator and the evaluation value.
14. An object recognition method comprising: generating for a plurality of different properties, by an information processing apparatus, property data that highlights a specific property based on target data acquired for an object; extracting for each piece of the property data, by the information processing apparatus, a discrimination-use feature amount used for discrimination of each piece of the property data; calculating for each piece of the property data, by the information processing apparatus, discrimination information used for discrimination of the property data based on the discrimination-use feature amount calculated for each piece of the property data; extracting for each piece of the property data, by the information processing apparatus, a reliability feature amount used for estimation of reliability of the discrimination information calculated for each piece of the property data; estimating for each piece of the property data, by the information processing apparatus, the reliability of the discrimination information based on the reliability feature amount calculated for each piece of the property data; generating, by the information processing apparatus, synthesized information acquired by synthesizing the discrimination information calculated for each piece of the property data and the reliability calculated for each piece of the property data; and performing, by the information processing apparatus, processing related to recognition of the object based on the synthesized information and generating a recognition result.
This invention relates to object recognition systems that improve accuracy by analyzing multiple properties of an object and synthesizing discrimination and reliability information. The method addresses challenges in object recognition where single-property analysis may lead to misclassification due to noise, occlusion, or ambiguous features. The system processes target data acquired from an object, such as images or sensor readings, to generate property data that emphasizes different properties (e.g., shape, texture, color). For each property, the system extracts discrimination-use feature amounts to distinguish between different property data and calculates discrimination information to classify the object based on these features. Additionally, reliability feature amounts are extracted to assess the confidence of the discrimination results. The system estimates reliability for each property's discrimination information and synthesizes this with the discrimination results into a unified output. The synthesized information combines the discriminative power of multiple properties with their respective reliability estimates, enabling more robust object recognition. The final recognition result is generated by processing this synthesized data, improving accuracy by leveraging complementary property information and reliability assessments.
15. The object recognition method according to claim 14 , further comprising: storing, by the information processing apparatus, a correspondence among learning-use discrimination-use feature amount, learning-use reliability feature amount, and a class as learning-use information; generating, by the information processing apparatus, the discriminator information based on the learning-use discrimination-use feature amount; generating, by the information processing apparatus, learning-use discriminator information based on the learning-use discrimination-use feature amount; an correctness or incorrectness evaluation unit that generates a correctness or incorrectness attached reliability feature amount by estimating a class based on the learning-use discrimination information and comparing the estimated class with a class corresponding to the learning-use discrimination-use feature amount, the correctness or incorrectness attached reliability feature amount being information including a result of correctness or incorrectness evaluation for each piece of the property data; and generating, by the information processing apparatus, the discriminator weight information based on a ratio of correctness to incorrectness grasped from the correctness or incorrectness attached reliability feature amount.
Unknown
March 30, 2021
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.